Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
XAI launched /goal inside Grok Build on June 22, a terminal coding agent mode that autonomously plans, executes, and verifies tasks without requiring step-by-step approvals. The feature uses a dual-model pipeline with Co…
Google account executive Yousuf Imran, who earned nearly $1 million last year, left the company in April to found AI startup Mangosteen Studio, citing FOMO around the AI boom and layoffs at Google. Business Insider publi…
Anthropic's Claude Code, a terminal tool marketed as a coding agent, can replace productivity apps by automating non-coding workflows. The article details six ways to use the tool for tasks like file management and data …
A developer introduced Modelis, an OpenAI-compatible gateway that charges a flat per-call price for bounded-output tasks like summarization, classification, and extraction, making it suitable for high-volume automations …
A developer released okf-skills, an open-source toolchain that implements the Open Knowledge Format (OKF) for Claude Code and other coding agents. OKF provides structured, versioned, and portable project knowledge that a…
A developer building technical dossiers warns against asking LLMs to verify their own hallucinated citations, as the model lacks access to a real database and suffers from self-evaluation bias. Instead, the developer imp…
A developer demonstrates that AI code review quality degrades when the reviewer is the same model that wrote the code, due to a documented self-preference bias. The solution involves using a separate reviewer from a diff…
A developer built a four-layer configuration system to enforce coding principles in Claude Code, moving beyond the suggestion-based CLAUDE.md file. The system includes a brain (CLAUDE.md), references, versioned skills, a…
Josef Šlerka released an open-source academic-writing kit for Claude Code and Cursor that guides an AI agent through producing a scholarly paper from raw insight to finished study while keeping the human as the author. T…
A developer building a personal knowledge base using Karpathy's LLM Wiki framework discovered that storing only declarative knowledge (facts and definitions) is insufficient for expert-level reasoning. By analyzing teach…
Andrew Ambrosino, lead of OpenAI's Codex desktop app, says nearly 100% of OpenAI employees now use Codex weekly and aims to make it the best desktop app ever. In a podcast, he discusses how AI flips product development, …
Researchers propose scaling parameter-efficient fine-tuning (PEFT) to support millions of personalized models on top of shared trillion-parameter foundation models. The study introduces three scaling axes—Scale Up, Scale…
New Zealand winger Elijah Just scored twice against Iran in the 2026 FIFA World Cup, becoming the first Motherwell player and first New Zealander to score at a World Cup. His performance has sparked transfer interest fro…
Coinbase CEO Brian Armstrong has switched the company to Chinese AI models GLM 5.2 and Kimi 2.7, using an automated routing system that selects the best model per task and price. Improved caching boosted hit rates from 5…
Parseflow, a new macOS app, extracts structured data from PDFs using a local AI model, ensuring 100% privacy with no cloud processing. Priced at €17.99 as a one-time payment, it offers unlimited usage and offline functio…
Nearly three-quarters of Dutch responses to stricter EU tobacco rules were AI-generated, with nine out of ten coming from a text generator by Philip Morris, according to an investigation by Pointer and NOS. The European …
Hugging Face's top-voted AI papers for June 28, 2026, reveal a shift from models that 'know how to answer' to systems that 'know how to act, remember, adapt, and generate context-aware content.' The roundup covers 10 not…
A developer shares hard-won solutions to two persistent LLM problems—hallucinations and memory loss—explaining that these issues are inherent to model architecture and offering practical fixes.
A developer upgraded a TrueNAS NAS with an AMD Ryzen 7 8700G APU and 96 GB DDR5 for local LLM inference, achieving 13-15 tok/s on gemma4, after rejecting pricier options like NVIDIA DGX Spark and Mac Studio. The build co…
A developer benchmarked speculative decoding using Qwen2.5-0.5B-Instruct as the draft model and Qwen2.5-1.5B-Instruct as the target model on a CPU. Across code, JSON, and story generation tasks, speculative decoding was …